maix.tensor

maix.tensor module

You can use maix.tensor to access this module with MaixPy
This module is generated from MaixPy and MaixCDK

Module

No module

Enum

DType

Tensor data types

item	describe
values	UINT8: INT8: UINT16: INT16: UINT32: INT32: FLOAT16: FLOAT32: FLOAT64: BOOL: DTYPE_MAX:

C++ defination code:

enum DType
        {
            UINT8 = 0,
            INT8,
            UINT16,
            INT16,
            UINT32,
            INT32,
            FLOAT16,
            FLOAT32,
            FLOAT64,
            BOOL,
            // STRING,
            // OBJECT,
            DTYPE_MAX
        }

Variable

dtype_size

Tensor data type size in bytes

item	description
attention	It's a copy of this variable in MaixPy, so change it in C++ (e.g. update var in hello function) will not take effect the var inMaixPy. So we add const for this var to avoid this mistake.
value	{ 1, // UINT8 1, // INT8 2, // UINT16 2, // INT16 4, // UINT32 4, // INT32 2, // FLOAT16 4, // FLOAT32 8, // FLOAT64 1, // BOOL // 1, // STRING // 1, // OBJECT 0 }
readonly	True

C++ defination code:

const std::vector<int> dtype_size = {
            1, // UINT8
            1, // INT8
            2, // UINT16
            2, // INT16
            4, // UINT32
            4, // INT32
            2, // FLOAT16
            4, // FLOAT32
            8, // FLOAT64
            1, // BOOL
            // 1, // STRING
            // 1, // OBJECT
            0
        }

dtype_name

Tensor data type name

item	description
value	{ "uint8", "int8", "uint16", "int16", "uint32", "int32", "float16", "float32", "float64", "bool", // "string", // "object", "invalid" }
readonly	True

C++ defination code:

const std::vector<std::string> dtype_name = {
            "uint8",
            "int8",
            "uint16",
            "int16",
            "uint32",
            "int32",
            "float16",
            "float32",
            "float64",
            "bool",
            // "string",
            // "object",
            "invalid"
        }

Function

tensor_from_numpy_float32

def tensor_from_numpy_float32(array: numpy.ndarray[numpy.float32], copy: bool = True) -> Tensor

float32 type numpy ndarray object to tensor.Tensor object.

item	description
param	array: numpy array object. copy: if true, will alloc new buffer and copy data, else will directly use array's data buffer, default true. Use this arg carefully, when set to false, ther array MUST keep alive until we don't use the return tensor of this func, or will cause program crash.
return	tensor.Tensor object.

C++ defination code:

tensor::Tensor *tensor_from_numpy_float32(py::array_t<float, py::array::c_style> array, bool copy = true)

tensor_from_numpy_uint8

def tensor_from_numpy_uint8(array: numpy.ndarray[numpy.uint8], copy: bool = True) -> Tensor

uint8 type numpy ndarray object to tensor.Tensor object.

item	description
param	array: numpy array object. copy: if true, will alloc new buffer and copy data, else will directly use array's data buffer, default true. Use this arg carefully, when set to false, ther array MUST keep alive until we don't use the return tensor of this func, or will cause program crash.
return	tensor.Tensor object.

C++ defination code:

tensor::Tensor *tensor_from_numpy_uint8(py::array_t<uint8_t, py::array::c_style> array, bool copy = true)

tensor_from_numpy_int8

def tensor_from_numpy_int8(array: numpy.ndarray[numpy.int8], copy: bool = True) -> Tensor

int8 type numpy ndarray object to tensor.Tensor object.

item	description
param	array: numpy array object. copy: if true, will alloc new buffer and copy data, else will directly use array's data buffer, default true. Use this arg carefully, when set to false, ther array MUST keep alive until we don't use the return tensor of this func, or will cause program crash.
return	tensor.Tensor object.

C++ defination code:

tensor::Tensor *tensor_from_numpy_int8(py::array_t<int8_t, py::array::c_style> array, bool copy = true)

tensor_to_numpy_float32

def tensor_to_numpy_float32(t: Tensor, copy: bool = True) -> numpy.ndarray[numpy.float32]

tensor.Tensor object to float32 type numpy ndarray object.

item	description
param	t: tensor.Tensor object. copy: Whether alloc new Tensor and copy data or not, if not copy, array object will directly use arg's data buffer, will faster but change array will affect arg's data, default true.
return	numpy array object

C++ defination code:

py::array_t<float, py::array::c_style> tensor_to_numpy_float32(tensor::Tensor *t, bool copy = true)

tensor_to_numpy_uint8

def tensor_to_numpy_uint8(t: Tensor, copy: bool = True) -> numpy.ndarray[numpy.uint8]

tensor.Tensor object to int8 type numpy ndarray object.

item	description
param	t: tensor.Tensor object. copy: Whether alloc new Tensor and copy data or not, if not copy, array object will directly use arg's data buffer, will faster but change array will affect arg's data, default true.
return	numpy array object

C++ defination code:

py::array_t<uint8_t, py::array::c_style> tensor_to_numpy_uint8(tensor::Tensor *t, bool copy = true)

tensor_to_numpy_int8

def tensor_to_numpy_int8(t: Tensor, copy: bool = True) -> numpy.ndarray[numpy.int8]

tensor.Tensor object to int8 type numpy ndarray object.

item	description
param	t: tensor.Tensor object. copy: Whether alloc new Tensor and copy data or not, if not copy, array object will directly use arg's data buffer, will faster but change array will affect arg's data, default true.
return	numpy array object

C++ defination code:

py::array_t<int8_t, py::array::c_style> tensor_to_numpy_int8(tensor::Tensor *t, bool copy = true)

Class

Tensor

Tensor class

C++ defination code:
class Tensor

init

def __init__(self, shape: list[int], dtype: DType) -> None

Tensor constructor

item	description
type	func
param	shape: tensor shape, a int list dtype: tensor element data type, see DType of this module
static	False

C++ defination code:

Tensor(std::vector<int> shape, tensor::DType dtype)

to_str

def to_str(self) -> str

To string

item	description
type	func
static	False

C++ defination code:
std::string to_str()

str

def __str__(self) -> str

To string

item	description
type	func
static	False

C++ defination code:
std::string __str__()

shape

def shape(self) -> list[int]

get tensor shape

item	description
type	func
return	tensor shape, a int list
static	False

C++ defination code:
std::vector<int> shape()

expand_dims

def expand_dims(self, axis: int) -> None

expand tensor shape

item	description
type	func
param	axis: axis to expand
static	False

C++ defination code:
void expand_dims(int axis)

reshape

def reshape(self, shape: list[int]) -> None

reshape tensor shape, if size not match, it will throw an err::Exception

item	description
type	func
param	shape: new shape
static	False

C++ defination code:
void reshape(std::vector<int> shape)

flatten

def flatten(self) -> None

Flatten tensor shape to 1D

item	description
type	func
static	False

C++ defination code:
void flatten()

dtype

def dtype(self) -> DType

get tensor data type

item	description
type	func
return	tensor data type, see DType of this module
static	False

C++ defination code:
tensor::DType  dtype()

to_float_list

def to_float_list(self) -> list[float]

get tensor data and return a list

item	description
type	func
return	list type data
static	False

C++ defination code:
std::valarray<float>* to_float_list()

argmax

def argmax(self, axis: int = 65535) -> Tensor

argmax of tensor

item	description
type	func
param	axis: By default, the index is into the flattened array, otherwise along the specified axis., wrong axis will throw an err::Exception
return	argmax result, you need to delete it after use in C++.
static	False

C++ defination code:
tensor::Tensor *argmax(int axis = 0xffff)

argmax1

def argmax1(self) -> int

argmax1, flattened data max index

item	description
type	func
return	argmax result, int type
static	False

C++ defination code:
int argmax1()

Tensors

C++ defination code:
class Tensors

init

def __init__(self) -> None

Constructor of Tensors

item	description
type	func
static	False

C++ defination code:
Tensors()

add_tensor

def add_tensor(self, key: str, tensor: Tensor, copy: bool, auto_delete: bool) -> None

Add tensor

item	description
type	func
static	False

C++ defination code:

void add_tensor(const std::string &key, tensor::Tensor *tensor, bool copy, bool auto_delete)

rm_tensor

def rm_tensor(self, key: str) -> None

Remove tensor

item	description
type	func
static	False

C++ defination code:
void rm_tensor(const std::string &key)

clear

def clear(self) -> None

Clear tensors

item	description
type	func
static	False

C++ defination code:
void clear()

get_tensor

def get_tensor(self, key: str) -> Tensor

Get tensor by key

item	description
type	func
static	False

C++ defination code:

tensor::Tensor &get_tensor(const std::string &key)

getitem

def __getitem__(self, key: str) -> Tensor

Operator []

item	description
type	func
static	False

C++ defination code:

tensor::Tensor &operator[](const std::string &key)

len

def __len__(self) -> int

Size

item	description
type	func
static	False

C++ defination code:
size_t size()

keys

def keys(self) -> list[str]

Get names

item	description
type	func
static	False

C++ defination code:
std::vector<std::string> keys()

tensors

Tensors data, dict type

item	description
type	var
static	False
readonly	False

C++ defination code:

std::map<std::string, tensor::Tensor*> tensors

err

image